Orange is taking a measured route toward autonomous network operations. The operator is not chasing autonomy for its own sake. Instead, it wants clear proof that agentic systems improve daily operations.
Philippe Ensarguet, Orange Fellow and VP of Cloud and Software Engineering, described this path recently. He framed it as four stages: assist, recommend, act and orchestrate. Each step gives systems more responsibility, but only after teams gain confidence.
“Assist first, earn trust, expand the scope,” Ensarguet said.
This approach matters because telecom networks carry high operational risk. A flashy autonomous demo can impress executives. Yet network teams need reliability, control and clear accountability. Ensarguet warned that skipping steps may create a “very, very impressive demo,” while leaving operations staff uncomfortable.
In practice, the first stage helps engineers work faster. AI can collect data, summarize incidents and suggest likely causes. The next stage recommends actions. Later, systems may act within approved boundaries. Only then can they coordinate across many network domains.
Orange has already tested this model in real network areas. At Orange OpenTech 2025, the company showed LiveCentriX. It uses a multi-agent model for 5G security monitoring. Multi-agent means several specialized software agents work together on one task.
The operator also sees value in cloud root-cause analysis. That helps teams find why failures happen. Other targets include RAN energy optimization, lifecycle management and incident response. These areas offer measurable gains when automation works well.
However, Orange is not selecting projects based only on technical appeal. Ensarguet highlighted a “business value first selection” process. The company uses a “high value scenario” method to find use cases. These use cases must scale across Orange affiliates.
“The use cases that scale for me are the ones chosen for business value, not for technical implementation.”
He added another direct message for telecom teams.
“We need to start with the problem and not with the technology.”
This discipline also extends to cost control. Agentic AI can consume large numbers of tokens. Tokens are small units of text processed by AI models. More tokens can mean higher operating costs. Yet total token volume tells only part of the story.
“I would treat tokens as unit cost, but measure them at a level of real outcome, token cost per assisted decision, per resolved case, per completed workflow, not total token conception.”
That view shifts the debate from usage to impact. Operators should measure whether AI changes a decision. Then they should link that change to an operational result. Examples include faster repair times, fewer outages and lower coordination effort.
Orange also wants stronger evidence before expanding automation. Phased rollouts can compare AI-assisted teams with control groups. This reduces the risk of confusing correlation with real cause.
The message for the telecom sector is clear. Autonomous networks will not arrive through one dramatic jump. They will emerge through controlled progress, trusted results and measured business value.
